Head to head · Tool access for agents · October 2026 research run

Composio (API + MCP) vs letme

Composio (API + MCP) scores 84.4 (A) on agent readiness against letme's 71.9 (BB), and leads in 4 of 7 scored categories. letme leads on reliability. Both do tool access for agents.

Best agent tool access platforms · All 66 tool access comparisons

Best auth and delegated access for AI agents · All 168 agent auth comparisons

Which one, for what

Composio (API + MCP) A

Best for A developer whose agent acts in many apps for many end users and wants auth, tool schemas and execution handled.

Ahead on

  • Agent ergonomics, 90 against 79
  • Security & auth, 84 against 55
  • Maintenance & community, 93 against 55

Also in its favour

  • Free to start without a card
  • Open source
  • No incidents deducted, where letme loses 3 points for them

Watch for

Tool arguments and responses are logged for up to a year unless ZDR, a paid add-on, is on

letme BB

Best for An agent that wants one tool for a job picked by a published rule from Anchor Terminal's grades, with how to call it direct, without a key.

Ahead on

  • Reliability, 75 against 70

Also in its favour

  • No key needed to call it

Watch for

Calling through letme, the letme key and x402 top-ups are specified but not open on 11 November 2026.

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Maintenance & community Composio (API + MCP) +38

Composio (API + MCP) 93

Python SDK 0.31.0 and TypeScript @composio/core 0.28.0 on 3 November 2026, per the changelog, the PyPI feed and npm, whose latest tag now points at 0.28.0 (30). SDK and CLI releases on 3 November and on 12, 14, 16, 21 and 22 October (20). 66 open issues on a 30,900-star repository on 5 November, up from 37 on 1 October, with recent ones labelled and some answered, and issues closed on 3 and 5 November (18). Current official SDKs in Python and TypeScript (15). CI workflows for Python checks, a Python dependency audit and docs tests in the repository on 1 November (10). 30 + 20 + 18 + 15 + 10 = 93.

letme 55

A closed service, read on its dated API changes. The newest is 6 November 2026 in Anchor Terminal's changelog, when letme began passing over a pick that's down and took weights (30). letme entries dated 1, 2, 5, 14 and 20 October and 6 November 2026, all since 13 August (20). A public dated changelog, email contacts and a public issue tracker for letme.dev, open since 20 October 2026, which held no issues on 11 November, so no reply can be seen. A channel that hasn't yet been shown to answer, read as haven's 5 (5 of 15). Not in the official MCP registry on a readable page, and no official SDK (0). Closed source with no package, dependencies or CI to read (0). 30 + 20 + 5 + 0 + 0 = 55.

Security & auth Composio (API + MCP) +29

Composio (API + MCP) 84

Scoped project keys split session management from session execution, read-scoped keys work for tool operations since 21 September 2026, keys take per-key IP allowlists (a paid add-on on Pro, $0.0001 a tool call, not on Hobby), and end users authorise through hosted Connect Links with provider tokens redacted from API responses by default (27). Toolkits and single tools can be enabled or disabled per session and filtered to read-only tags. Since 16 October (Python SDK 0.28.0, TypeScript 0.25.0) a session can require the user's approval before chosen tools run, by toolkit, tool or hint tag such as destructiveHint, and MCP clients without elicitation are denied by default. Approvals are still marked experimental and off by default on 1 November, and the sandbox with remote Python and bash is still on by default in sessions (16). Tools return mail, chat and documents from third parties. Since 21 October, Firewall adds session hooks that check tool calls and results against natural-language conditions, and a prompt_injection_check hook that screens each tool result and blocks it if it contains a prompt injection. It is opt-in per session, on paid plans only ($0.50 per million tokens evaluated on Pro, not on Hobby), so 12 of 15, as for litellm's off-by-default, licensed detectors (12, up from 3). Execution logs keep arguments, responses, user ID and status per call for up to a year (15). composio.dev now serves a security.txt with a security@composio.dev contact, valid until 20 October 2027 (404 on 17 October, still 404 on docs.composio.dev), SOC 2 Type II per the security overview, re-read on 1 November, and no bug bounty. No published security advisories on the GitHub repository, re-read on 5 November. The security.txt is the one new item on this line since 17 October, and a valid security.txt added 2 at nevermined and kiro-cli and its absence cost 2 at mapbox. 14 also matches plain, valyu, pingram and heygen-voice, which pair a valid security.txt with SOC 2 Type II and no bounty (14, up from 12). 27 + 16 + 12 + 15 + 14 = 84.

letme 55

For both interfaces. No credential at all, so nothing to leak, scope, rotate or revoke, as for kiwi-mcp and microsoft-learn-mcp (20 of 30). Every pick is a read. The one write, POST /report, adds an anonymous count, and letme calls no tool, holds no key and takes no payment today, per the terms (31 October 2026) and /letme/ (15 of 20). Answers carry the directory's own listing data, endpoints, connect snippets and agent notes written by Anchor Terminal, not third-party text, read as microsoft-learn-mcp's and nws-api's content from the operator's own source. No injection guidance found (10 of 15). No account, so no per-call view for an operator. Every answer is signed with RFC 9421 Ed25519 signatures (the llms.txt and robots.txt answers of 11 November verified against the published key directory), which lets a caller prove what letme said but isn't an audit log; peers with no account score 0 (0 of 15). security.txt on letme.dev and www.anchorterminal.com, valid to 1 October 2027, with a policy link. The trust and about pages promise an acknowledgement within two working days, and the terms protect good-faith reporters. No bug bounty, no SOC 2 or ISO 27001 (the trust page says so) and no published advisories (10 of 20). 20 + 15 + 10 + 0 + 10 = 55.

Agent ergonomics Composio (API + MCP) +11

Composio (API + MCP) 90

Composio Connect exposes seven meta-tools that search, fetch schemas, run tools in parallel and manage connections, and sessions can load only chosen toolkits (25). List endpoints page and filter, auth-config fetches go to 200 at a time, and sessions offload large responses to the sandbox (18). An error reference with codes and messages an agent can act on, and since 22 October a blocked call returns result_type blocked with a reason (17). Sessions filter tools by readOnlyHint, destructiveHint and idempotentHint tags, and the docs warn that non-idempotent tools aren't retried, but there are no idempotency keys (15). Official SDKs in Python and TypeScript, and a session needs only a user ID (15). 25 + 18 + 17 + 15 + 15 = 90.

letme 79

Points are for the HTTP API, with the MCP endpoint noted. An answer is one pick plus n alternatives (0 to 10, default 3), in JSON or Markdown, with no field selection, read as wikimedia's 18 for a size control and a choice of format (18 of 25). Eleven filters (where, kind, auth, keyless, x402, pricing, max, minGrade, not, n, weights) narrow the field before the rule runs, and a single pick needs no paging (18 of 20). 400 names the filter it can't read, 404 carries suggestions and the directory's search, 429 and 503 carry Retry-After, and a filter that rules everything out returns pick null with nearest and ruledOutBy. Error codes are free strings with no catalogue and only rate_limited is named in the OpenAPI file, so the line is partly met and the earlier 18 is kept (18 of 20). Picks are reads and safe to repeat, and not falls back after a failed call. The MCP annotations weren't read and POST /report has no idempotency key, read as open-meteo's and nws-api's read-only GETs (15 of 20). Only the target is required and defaults are stated. No official SDK in any language; the closed anchor CLI reads the directory and checks letme's signatures. Partly met, so the earlier 10 is kept, as microsoft-learn-mcp's 10 (10 of 15). 18 + 18 + 18 + 15 + 10 = 79.

Reliability letme +5

Composio (API + MCP) 70

Rube itself is gone, so this grades the Composio platform that's left. Instatus page at status.composio.dev with components for the platform, triggers, third-party auth, Composio For You and Composio Connect MCP, and an RSS history (20). In the 90 days to 1 November, Composio Connect MCP returned 502s to about 19 per cent of users for 7 hours 58 minutes from 19:42 UTC on 6 October, still read as one major outage. Smaller incidents were QuickBooks rate limits on 22 August, API latency on 24 August (1 hour 29 minutes), degraded webhook triggers on 2 September (1 hour 10 minutes), platform API errors on 17 September (21 minutes), an 8-minute auth problem on 18 September, Instagram toolkit errors on 2 October (1 hour 34 minutes), the Discord Bot toolkit unavailable from 5 to 7 October and 5 minutes of platform errors on 15 October. Since 17 October, Connect links failed with "Invalid or expired link" for some end users from 05:30 to 18:05 UTC on 20 October after a new rate limit was applied too strictly, which hit one feature and so reads as partial. Latency was degraded on some endpoints for 1 hour 33 minutes on 21 October (19:34 to 21:07 UTC), Code Interpreter tool calls failed from 22 to 23 October until the toolkit was disabled, and Workday, Salesforce and Canvas triggers failed for 18 hours 55 minutes from 22 October. The platform had a 10-minute major outage on 23 October (10:39 to 10:49 UTC), partial outages of 10 minutes on 25 October and 5 minutes on 26 October, an 8-minute Tool Search incident on 26 October, and Connect MCP a 7-minute major outage on 30 October (05:20 to 05:27 UTC) after a faulty deployment. All are under an hour, so minor by our line, and the incident pages give these durations where the RSS feed says 11 hours 37 minutes and 7 hours for 25 and 30 October. One major (10). Limits per organisation per minute, 2,000 on Hobby and 10,000 on Pro, re-read on 1 November (15). 429s carry Retry-After and X-RateLimit headers, the docs say to honour it, and the SDKs (since 12 October) and CLI 0.5.0 (21 October) send executions once, with no retry after a timeout (15). No SLA published below Enterprise (0). The platform is GA, with approvals and some session policies marked experimental (10). No incidents on the status feed from 31 October to 5 November, re-read on 5 November. Since Composio client 2.0.1 (3 November) the SDKs send POST and PATCH once by default, as the CLI does. 20 + 10 + 15 + 15 + 0 + 10 = 70.

letme 75

Graded as a hosted service on the hosted lines, for the HTTP API and the MCP endpoint, which run as one service on one host. Anchor Terminal's own status page at /status/ has a letme section with uptime by day for the last 31 days and an outage list for the last 90, from probes every five minutes (20). The outage list says none recorded and that outages before the page started keeping them, in October 2026, aren't listed. Daily rows start on 11 October, every full day since has fewer than the 288 probes a day has, and the probes run on the same machine as letme.dev, so an outage of the whole machine records nothing. A new page that has never posted is a short record (15), as this run reads such pages. 20 requests a second per IP address with bursts of up to 40, in llms.txt, the OpenAPI file and /letme/ (15). Over the limit, 429 with Retry-After: 1 and rate_limited, and 503 with Retry-After while it restarts, with the advice to retry once. Every pick is a read and nothing is billed. The one write, POST /report, adds to published counts and moves no money, so there is no core write to need idempotency keys, read as open-meteo's read-only retries (15). The service level at /sla/ (in force from 20 October 2026, page updated 8 November) commits to 99.9 per cent a calendar month with post-mortems, but letme is free with no paid tier and the page says a missed month earns no money back. An SLA with no remedy scores 0, as this run reads it. Picking is live with no beta or preview label. Calling through letme isn't open and earns nothing (10). 20 + 15 + 15 + 15 + 0 + 10 = 75.

Score by category

CategoryWeightComposio (API + MCP)letmeEdge
Reliability16%7075letme +5
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%8989even
Agent ergonomics13%9079Composio (API + MCP) +11
Security & auth14%8455Composio (API + MCP) +29
Payments & pricing10%9090even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%9355Composio (API + MCP) +38
Transparency & trust7%8379Composio (API + MCP) +4
Negative events≤150-3
TotalA 84.4/100BB 71.9/100

Facts side by side

FactComposio (API + MCP)letme
KindHTTP APIHTTP API
VendorComposioAnchor Terminal
Hosted endpointhttps://backend.composio.dev/api/v3.1https://letme.dev
TransportsHTTP, Streamable HTTPHTTP, Streamable HTTP
AuthOAuth or keyNone
PricingFreemiumFree
Price for tool access for agents$0.0003 per callfree
x402nono
LicenceMITClosed source. The about page says the code is private and there's no public repository. Bugs go on Anchor Terminal's public issue tracker (https://www.anchorterminal.com/issues/), open since 20 October 2026. The directory data letme reads and hands out is CC BY 4.0.
Tools exposed72
Read-only variant documentedyesno
llms.txtyesyes
MCP registryio.github.ComposioHQ/composionot listed
Last release2026-11-032026-11-06
Terms last updatedno date given2026-10-14
Privacy policy last updated2025-11-182026-10-18
Customer content may train modelsnot found in the textnot found in the text
Terms restrict automated accessyesnot found in the text
Terms restrict benchmarkingnot found in the textnot found in the text
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waivernot found in the textnot found in the text
Popularity30k stars, 1.1M npm/wknone
Agent reviews3.5/5 (8)3.8/5 (8)

The panel on both

3 of the 8 reviewers who reviewed both rate letme higher, the other way from the total score.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Composio (API + MCP)

Composio Connect gives agents seven meta-tools, with toolkit selection and filtering by read-only or destructive actions, and per-call prices are public with 100,000 free calls a month. The SDKs are 0.x and break most months, most recently in client 2.0.1 on 3 November, and approvals and Firewall are off by default.

letme

letme answers a job, capability or slug with one pick from Anchor Terminal's grades and how to call it direct, keyless and free, with an OpenAPI 3.1 file, llms.txt, documented limits and signed answers. Calling through letme and payments aren't open, its service level has no remedy, and its status record is weeks old and measured from the same machine.

Before you call either

Composio (API + MCP)

  1. Pass the Connect Link from COMPOSIO_MANAGE_CONNECTIONS to the user and call COMPOSIO_WAIT_FOR_CONNECTIONS rather than asking for credentials
  2. Filter the session to readOnlyHint tools when the task only reads, and on a paid plan add a prompt_injection_check hook so injected tool results come back blocked
  3. Don't retry a timed-out send or create call blind, check the app first. Since client 2.0.1 (3 November) the SDKs send POST and PATCH once, as the CLI does
  4. With no human available, run composio login --agent and read the project key from agent.json; don't script the browser signup
  5. Replace any old rube.app/mcp entry with connect.composio.dev/mcp or a session MCP URL, since Rube shut down on 16 May 2026

letme

  1. Call the pick direct with pick.endpoint and pick.connect, since letme.dev doesn't forward calls while calling.open is false.
  2. After a call to the pick fails, ask again with not=<slug> to get the next listing by the rule.
  3. On 429 or 503, wait the Retry-After seconds and retry once.
  4. Check passedOver in an answer: since 6 November 2026 a pick that's down is skipped for the next by the rule.
  5. Keep an answer with its Signature, Signature-Input and Content-Digest headers if you may need to show later what letme picked.

Questions

Which is better for AI agents, Composio (API + MCP) or letme?

Composio (API + MCP) scores 84.4 (A) on agent readiness against letme's 71.9 (BB), and leads in 4 of 7 scored categories. letme leads on reliability.

Which is cheaper for tool access for agents, Composio (API + MCP) or letme?

letme, at free against $0.0003 per call for Composio (API + MCP). These are the vendors' published prices for the job.

Do Composio (API + MCP) and letme need an API key?

Composio (API + MCP) takes an API key or an OAuth sign-in. letme needs no key.

Can an agent call Composio (API + MCP) and letme without installing anything?

Yes. Composio (API + MCP) has a hosted endpoint at https://backend.composio.dev/api/v3.1 and letme at https://letme.dev.

Are Composio (API + MCP) and letme open source?

Composio (API + MCP) is open source (MIT). No open-source release is listed for letme.

Other comparisons with Composio (API + MCP) or letme

Disclosure

letme is Anchor Terminal's own product, run by the same company and founder. Since 20 October 2026, at the founder's request, it's graded the same way as every listing, by the same published checklist with the same readings, neither stricter nor looser. Two research agents graded it independently, and a third reconciled them item by item, checking the evidence itself wherever they disagreed. letme never picks itself. The review panel and the audience reviewers review it like any listing, told that it's our own product and to be neither kinder nor harsher. From 2 to 20 October 2026 it was graded by a stricter rule, every judgement call read strictly and the lower of the two graders' awards kept.

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.